Support for ARM ECMDQs

[Sukadev Bhattiprolu]

>On 10/14/26 15:07, Sukadev Bhattiprolu wrote:
>> Hi,
>>
>> We are trying to understand the direction or ongoing efforts to 
>> support enhanced command queues (ECMDQs) in Linux.
>>
>> Earlier version of patches [1] seem to have stalled due to lack of 
>> real-world workload data.
>>
>> Also, any recommendations on policies to distribute available eCMDQs - 
>> one per CPU as in [1] or, round-robin when # of eCMDQs are less than # of CPUs or others?
>>

> I'd expect a static spread over all CPUs to improve caching of the cmdq memory structures.

>> Thanks,
>> Sukadev
>>
>> [1] https://lkml.indiana.edu/hypermail/linux/kernel/2404.3/07988.html

>As you can see in that thread, the maintainers wanted to see results from real-world workloads (which never appeared).

>And I think that the preemption disabling scheme in that series can be improved (dropped).

Any recommendations on the design - is the approach in this patch set such as associating a set of queues with arm_smmu_device the preferred ? Or try to adopt to the cmdv interface?

Thanks,
Sukadev